AI Person Swap

AI person swap for video — replace the person in any video with a new person from a single reference photo. Motion, framing, and audio of the source video are preserved.

"Replace the person in the source video with the uploaded reference person. Preserve the original motion and framing."

Input Image

Input Image
Input Image

Input Video

Input Video

Result

Result
Orientation: video

Describe your scene and generate a video in seconds

8,500+ videos generated this month

📄 About AI Person Swap
Key Features
Replace the person in any video with a single reference photo — full-body identity swap, not just a face overlay.
Motion, framing, camera path, and timing of the source video are preserved frame-by-frame.
Two orientation modes — 'Video Match' (up to 30s) for full motion, 'Image Match' (up to 10s) for camera-following swaps.
Optional text prompt for fine-tuning lighting, outfit, mood, or environment without breaking motion fidelity.
Keeps the original audio track from the source video by default — dialogue, music, and ambience stay synced.
Powered by Kling v3 motion control, the same engine used for professional dance and gesture transfer.
No signup gate, no installs, commercial-use rights included on every paid generation.
💡 Use Cases
Swap the lead in a music video, dance reel, or short-form social clip without a reshoot.
Generate localized advertising variants with different talent for different regions or markets.
Re-cast tutorial, presenter, or explainer videos with a new on-camera identity.
Pre-visualize casting choices for filmmakers, ad agencies, or content studios before booking actors.
Create response, reaction, or remix content for TikTok, Instagram, and YouTube with a new identity.
Build personalized video gifts by inserting a friend, family member, or yourself into an existing clip.
Produce branded UGC-style ads at scale by swapping a single reference photo per creative variant.
🎯 Best For
🎯 Marketers, editors, music-video creators, ad studios, and indie filmmakers who need to replace a person in a video without a reshoot, rotoscoping, or VFX pipeline.
👍 Pros
Full-body identity swap, not a flat face overlay — body, hair, and silhouette transfer naturally.
Original audio track is preserved by default, so dialogue and music stay in sync.
Only needs one reference photo of the new person — no training, no LoRA, no dataset.
Two orientation modes adapt to either complex motion or aggressive camera moves.
Cents per swap on pay-as-you-go credits — no subscription required.
Commercial-use rights included on every paid generation.
⚠️ Considerations
Reference photo must show a clear face and body — sunglasses, heavy occlusion, or tiny faces hurt fidelity.
Source clips longer than 30 seconds need to be split before swapping.
Scenes with multiple people may not isolate the intended subject cleanly — single-person shots work best.
Extreme camera shake, rapid cuts, or motion blur can reduce swap quality.
📚 How to Use AI Person Swap
1
Pick the source video that contains the person you want to replace — keep it under 30 seconds and trim heavy cuts.
2
Pick a clean reference photo of the new person — full or upper body visible, clear face, good lighting.
3
Upload both files: the new person as the image input, the source video as the video input.
4
Choose orientation mode — 'Video Match' for full motion transfer (up to 30s), 'Image Match' for camera-following swaps (up to 10s).
5
(Optional) Add a short scene prompt to influence lighting, outfit, or environment — leave blank for a clean default swap.
6
Decide whether to keep the original audio (default ON) — disable only if you plan to add custom audio in post.
7
Submit, wait 60–150 seconds for the swap, then download the output as MP4. The original is untouched — re-run with a new photo to spin variants.
💡 Pro Tips for AI Person Swap
Use One Clean Reference Photo, Not a Collage AI Person Swap reads identity from a single image, not a folder. Pick the highest-quality photo where the new person's face and body are both clearly visible — no sunglasses, no extreme angles, no group shots. If your only photos are from far away, crop tight on the person and upscale before uploading. A clean single-person reference outperforms a 'best photo we have' compromise every time.
Match Source Video Framing to Reference Photo Style If your reference photo is a head-and-shoulders portrait, pick a source video where the person stays in upper-body framing — full-body source clips will force the model to invent legs and lower-body proportions, which can drift. If your reference is full-body, the model has more to work with and full-body source clips swap cleanly. Match the reference framing to the source framing for best fidelity.
Choose 'Video Match' for Motion-Heavy Clips, 'Image Match' for Camera Moves 'Video Match' (up to 30s) gives the new person the source video's full motion arc — perfect for dance, gestures, walking, sports. 'Image Match' (up to 10s) anchors the new person to the orientation of their reference photo and lets the source camera move around them — ideal when the source has aggressive pans or zooms you don't want the new person aping. When in doubt, start with 'Video Match'.
Use the Prompt for Lighting & Outfit, Not for Motion The motion is locked to the source video, so don't write motion instructions in the prompt. Instead, use it to nudge lighting ('warm sunset lighting'), outfit ('white t-shirt and jeans'), or environment ('studio backdrop'). Keep the prompt under two short sentences — long prompts can fight the source motion and produce drift.
Trim the Source Clip Before Swapping If your source video has multiple shots or cuts, trim down to a single continuous take before swapping. The model treats the whole input as one continuous person — cuts to a different angle or a different person can produce identity drift mid-output. Single take = clean swap.
Spin Variants by Changing Only the Reference Photo Once you have a source clip that swaps cleanly, treat it as a template. Re-run the swap with different reference photos to spin localized ad variants, casting tests, or multi-talent series — the source video doesn't change, only the identity. This is one of the fastest ways to produce ad creative variations without re-shooting. Pair with JAI Music Clip Generator to add a new soundtrack per variant.
Frequently Asked Questions
Face swap tools paste a face onto an existing body — the body, hair, and silhouette of the original person stay in the video. AI Person Swap rebuilds the person from your reference photo and re-renders them performing the source motion, so hair, body shape, posture, and outfit silhouette all match the new identity instead of the original.
It works on most single-person clips up to 30 seconds with stable framing. The source video should show the subject clearly — extreme camera shake, rapid cuts, multiple overlapping people, or heavy motion blur will reduce swap quality. For longer videos, split them into ≤30s segments and run each separately.
Yes by default. 'Keep original sound' is ON, so dialogue, music, and ambient sound from the source video are preserved on the swapped output. Disable it only if you plan to add custom audio in post-production or replace the soundtrack.
Reference photo: a single high-quality image showing the new person — full or upper body, clear face, good lighting. Source video: under 30s for 'Video Match' mode, under 10s for 'Image Match' mode. For longer source footage, split into segments before swapping.
Yes. All paid generations on JAI Portal come with full commercial-use rights, including AI Person Swap outputs. Make sure your source video and reference photo are either original creations or properly licensed — JAI Portal grants commercial rights on its output, not on third-party content you upload.
Output quality sits between traditional face-swap tools and a real re-shoot. Because the model rebuilds the full body from the reference photo (rather than overlaying a face on the source body), the result reads as a fresh take with a new performer, not a sticker on top of the old performer. Quality is best when the reference photo is high resolution, the source clip is stable, and the framing of both matches. Heavy compression, multiple people in frame, or extreme camera motion can reveal AI artifacts — for those edge cases, plan to do a light cleanup pass in your editor.
AI Person Swap is built for single-person clips. If your source has multiple people, the model will target one subject (usually the most prominent) and may blend identities on secondary people. For multi-person scenes, isolate each person into a separate clip first (using masking in your editor), swap each separately, then composite the swaps back together. A simpler alternative for face-only changes is AI Video Body Swap or Video Head Swap, which are tuned for different swap scenarios.
AI Person Swap uses duration-based pricing — a 5-second swap costs roughly a fifth of a 25-second swap. Exact credit cost is displayed before you submit each generation, so you can dial duration up or down to match your budget. The 'Image Match' mode (up to 10s) is cheaper than 'Video Match' (up to 30s) since it processes a shorter clip. Pay-as-you-go means no subscription — top up once, run as many swaps as you need.
Output is MP4, H.264 encoded, optimized for web and social platforms. Aspect ratio matches the source video (16:9, 9:16, 1:1, or 4:5) so vertical TikTok/Reels clips stay vertical and horizontal YouTube clips stay horizontal. Resolution typically matches the source video; for sub-1080p sources, the output may not be sharper than the input. Frame rate matches the source (24 or 30 fps in most cases). If you need 4K outputs, upscale the result with Topaz Video Upscaler after the swap completes.
Yes — every paid generation on JAI Portal grants full commercial-use rights, including for ads, branded content, music videos, and client deliverables. The catch is that JAI Portal grants rights on the AI output, not on third-party content you upload. If your reference photo or source video belongs to someone else, you need their permission (or a license) before using the swap commercially. For consent-required scenarios (using a real person's likeness), always get written permission — AI cannot replace legal release forms.
If results drift, work the inputs first before re-rolling. (1) Try a higher-quality, better-lit reference photo with the face clearly visible. (2) Trim the source video to a single continuous shot — cuts and rapid scene changes are the #1 source of identity drift. (3) Switch orientation modes — sometimes 'Image Match' produces cleaner results than 'Video Match' for camera-heavy footage, and vice versa. (4) Add a short prompt nudging lighting or outfit to match the source. If you've tried all four and still aren't happy, swap to a tighter framing on both inputs (head-and-shoulders only) — the model has fewer degrees of freedom to drift on. For final cleanup, run the result through your editor to color-match against any unaffected clips around it.
⚖️ How AI Person Swap Compares
AI Person Swap is JAI Portal's flagship tool when you need to ai video replace person — the full identity (face, hair, body, silhouette) is rebuilt from one reference photo while the source video's motion, framing, and audio are preserved. Compared to AI Video Body Swap, which is tuned for keeping a face and changing the body underneath, Person Swap goes the other direction — change the whole person, keep the performance. Against Video Head Swap, which only replaces the head/face region, Person Swap is the right call when you want the entire silhouette to match the new identity (different hair length, different build, different outfit). For traditional face-overlay workflows where you want maximum face fidelity but the original body to stay intact, the head/face tools remain the right choice. If your project is a single still image rather than a video, use AI Image Body Swap or AI Image Head Swap instead. Pick AI Person Swap when you need to swap person in video at the level of a re-shoot — different lead, same scene — without the cost, time, or coordination of an actual re-shoot. Start with a free account, test a 5-second clip, then scale up to longer durations once you've dialed in the reference photo and framing.

More Video Generation Models